I don't think I showed this piece of evidence on this forum yet. It shows how the three important draconic aliased periods for QBO are discovered via machine learning.
Fig 1: Eureqa results when applied to a QBO time-series 2nd derivative. The lowest error solution (shown in reverse highlighted blue) feature as strongest sinusoidal factors the three seasonally aliased Draconic harmonics.
Fig. 2 : Alignment of the three strongest sinusoidal periods found via machine learning with those predicted via seasonally reinforced Draconic lunar tidal periods.
More details here: http://contextearth.com/2016/02/13/qbo-model-validation/